| name | x-tweet |
| description | Generate tweets that match the voice profile and resonate with your target audience. Tweets should feel like authentic observations from an experienced insider, not marketing content. |
X Tweet Generator
Overview
Generate 3 tweets per day that sound like a clear-eyed insider sharing observations, not doing content marketing. Voice, audience, and topics are defined by your context profiles.
Before Writing
Step 1: Read Context
-
Read context profiles in /context/:
voice.json — How to sound
audience.json — Who you're writing for
- Any business profile JSONs — Business context
-
Check engagement data in /knowledge/engagement/ for what's working
-
Check tweet history — Read /knowledge/content/tweet_history.json for previously written/posted tweets. Do NOT repeat the same topic or angle unless the user explicitly asks. If a topic was already covered, find a different angle or skip it entirely.
Step 2: Scrape Engagement (MANDATORY — run BEFORE research)
- Scrape your profile — Run
scrape_profile.py to get latest engagement data on recent tweets. Update engagement_log.json. Review what's working (likes, views, replies) and what's not. Use this to inform today's tweet strategy.
Step 3: Research (MANDATORY — use researcher-agent for all 3 in parallel)
- Current tech news — Search for today's top tech headlines (AI, startups, VC, big tech). Get specific names, numbers, details.
- Industry Twitter trends — Search for what's trending among your target audience, industry news, exits, funding, layoffs, controversies.
- Topics you care about — Search for topics relevant to your voice profile and business profile (defined in context profiles).
Step 4: Write
- Mix news-reactive and evergreen tweets. At least 1 tweet should react to a specific current event/news item. The others can be original pattern observations, but should be informed by what's happening now — not generic.
Tweet Types (Rotate Daily)
1. Pattern Observation (Primary — 40% of tweets)
Spot something in your industry and name the pattern.
- Start with a specific detail or observation
- Build to a broader insight
- Let the pattern speak for itself
2. Contrarian / Anti-Hype (25% of tweets)
Challenge conventional wisdom, call out spin, cut through noise.
- Mock discourse fatigue
- Question what everyone accepts
- Skeptical but not cynical
3. Founder/Insider Insight (20% of tweets)
Share practical wisdom from experience.
- Real operational insight, not motivational fluff
- Specific enough to be useful
- From experience, not theory
4. Cultural Commentary (15% of tweets)
Industry culture, work life, ecosystem observations.
- Natural language mixing if bilingual
- Pop culture timestamps
- Self-deprecating mundane contrasts
Voice Rules
ALWAYS
- Sound like someone who's been in the room
- Be a pattern spotter, not a preacher
- Be skeptical but not bitter
- Use simple language for sophisticated observations
- Short paragraphs, breathing room
- Trust the reader to connect the dots
- Follow voice.json for specific voice patterns
- When citing specific data/statistics, always attach the source with a link at the end of the tweet (e.g. "מקור: bloomberg.com/news/articles/...")
NEVER
- Management consultant speak
- Breathless hype ("game-changing", "revolutionary")
- Motivational speaker energy
- Corporate press release tone
- Humble-bragging
- Excessive emojis
- Thread announcements
- Generic advice
Signature Phrases
Use sparingly — pull from voice.json signature_concepts and colloquialisms.
Twitter-Specific Patterns
- Brevity: Cut connective tissue, not substance
- Single-line punch: One powerful observation, full stop
- Setup → Evidence: First tweet poses observation, thread continues with proof
Language
Defined by your voice.json. Supports:
- Primary language tweets
- Secondary language tweets for broader reach (1 in 3)
- Natural code-switching between languages
Length
- Ideal: 1-3 sentences (under 280 chars)
- Max: 4-5 sentences for thread starters
- Short and punchy beats long and complete
Daily Generation
Generate 3 tweets with variety across languages and tweet types.
Each tweet should be independent (not a thread) unless specifically requested.
Step 5: Humanize (MANDATORY — run AFTER writing)
- Run humanizer — Apply the
/humanizer skill to all tweets before presenting. Remove AI-isms, inject personality, ensure they sound like a real person wrote them.
Quality Checklist
Before delivering:
Tweet History
After presenting tweets to the user, always update /knowledge/content/tweet_history.json with each tweet's date, topic, angle, status, and text preview. This prevents repeating the same topics/angles across sessions.
Engagement Learning
Check /knowledge/engagement/engagement_log.json for data on what works.
Patterns to optimize for:
- Which tweet types get most engagement
- Which topics resonate
- Language performance differences
- Time-of-day patterns
- What language/phrases trigger shares